Things you can do to optimize your neural network
1. Normalize data
2. Change activation function
3. Change the loss function
4. Add dropout if you are overfitting
What can you do to make your model more "general purpose"?
Adding a dropout layer removes the "crutches" that a model might be using and forces it
to make connections with different neurons. It makes the network less reliant on certain neurons to make
final assessments.
Obviously this needs to happen with multiple epochs. Maybe its the case that it needs more epochs the more
dropout you introduce.
Pretrained models for fine-tuning
https://keras.io/api/applications/